Murder & Medicine

Adjusting US homicide rates for improvements in trauma care, 1900–2020.

Chart 1 · Normalized to 1964 = 1.0: Assault, Observed Homicide & Adjusted Homicide
Sources: Harris et al. 2002 (1964-1999); FBI UCR / BJS (2000-2020); CCJ lethality brief (2025). Adjustment uses firearm-specific lethality ratio from Harris.
Chart 2 · Lethality Curves: Firearm Lethality (UCR, 1964–2020) & Abdominal GSW Mortality (Clinical, 1900–2020)
Firearm lethality: Harris et al. weapon-specific data (1964-1999); Braga & Cook 2018; CCJ. Abdominal GSW mortality: interpolated from cited clinical anchor points (see methodology below).
Chart 3 · Normalized to 1964 = 1.0: Observed Homicide, Adjusted by Abdominal GSW Survival (1900–2020), vs. Adjusted by Firearm Lethality (1964–2020)
Abdominal GSW mortality anchor points from clinical/military literature, linearly interpolated. Homicide rates from Vital Statistics (1900–1960) and UCR (1960+). See methodology below.

⚠ Methodology & Caveats

Abdominal GSW mortality anchors: Cited anchor points with linear interpolation between them: 87% (1900, based on Civil War mortality per Barr et al. 2023, which persisted through the Spanish-American and Boer Wars); 30% (1960, Wilson & Sherman 1961, 494 civilian cases); 12% (1988, Demetriades et al. 1988, 300 cases at urban Level I center); 9.5% (1995 consensus, multiple series); 11% (2010, ACS TQIP national database, 16,866 patients; higher than 1990s due to worsening injury severity); 8% (2020, modern Level I standard). Corroborating: WWI forward hospitals reduced military abdominal mortality from 65% to 45%; Giacopassi et al. 1992 (Memphis, 25-year intervals) shows overall assault lethality 11.4% (1935) → 5.5% (1960) → 3.2% (1985).

Adjustment formula: adjusted(t) = observed(t) × mort(1964) / mort(t). All series normalized to 1964 = 1.0. This is an upper bound on the true adjustment because not all homicide victims are medically saveable: head shots (about 11% of assault GSWs, roughly 90% fatal regardless of era) and DOA cases are essentially invariant. The true adjustment is probably 60–80% of the displayed values for recent decades.

Other limitations: (1) Pre-1960 anchor points are from military data, not systematic civilian surveillance. (2) 1900–1910 homicide rates are from "registration states" only and undercount the violent South. (3) The abdominal series cannot capture changes in injury composition (caliber, wounds per victim); see Webster et al. 1992 on the crack-era shift to semiautomatics. (4) Linear interpolation between sparse anchors is crude. This chart is illustrative, not definitive.

Data Sources & Notes

Harris et al. (2002), "Murder and Medicine": The foundational study. Firearm-specific lethality dropped 65% from 1964 (15.5%) to 1999 (5.4%). Only 1.2% of the overall lethality decline was attributable to weapon mix shifts; 98.8% was genuine within-weapon improvement. Motor vehicle crash lethality dropped 67% over the same period, supporting the medical explanation. Counties with hospitals had 11–24% lower lethality; regionalized trauma systems added 16% further reduction. Harris data starts at 1964; the firearm lethality series in Charts 1 and 3 begins there.

Giacopassi et al. (1992), Memphis study: The only study reaching back before 1960. Memphis police homicide files at 25-year intervals: 1935 lethality 11.4%, 1960 lethality 5.5%, 1985 lethality 3.2%. The 1935→1960 drop (roughly 2×) is driven by penicillin, blood banking, and hospital proliferation.

Assault denominator problems: The Harris firearm lethality ratio uses firearm aggravated assaults as its denominator. Under UCR rules, any assault involving a firearm is classified as aggravated by definition, making it resistant to the felony-to-misdemeanor reclassification documented by Eterno and Silverman (2012) under CompStat. However, the firearm assault count is still subject to changes in reporting completeness. O'Brien (2003) argues long-term UCR assault increases reflect reporting expansion, not actual violence. The abdominal GSW survival approach in Chart 3 sidesteps this entirely: it uses clinical mortality, not crime reporting.

Why gunshot hospitalizations don't solve it cleanly: CDC explicitly warns against using NEISS-AIP for firearm injury prevalence due to small, geographically unrepresentative hospital samples. Nonfatal data starts only 2001 (partially 1993). The best available proxy remains weapon-specific lethality ratios supplemented by clinical case fatality studies.